Machine Vision and Applications (2018) 29:393–403
https://doi.org/10.1007/s00138-017-0890-y
ORIGINAL PAPER
Classify vehicles in traffic scene images with deformable
part-based models
Shuang Bai
1
· Zhenyao Liu
1
· Chang Yao
2
Received: 10 March 2017 / Revised: 14 October 2017 / Accepted: 22 October 2017 / Published online: 18 November 2017
© Springer-Verlag GmbH Germany, part of Springer Nature 2017
Abstract Vehicle classification is an important and chal-
lenging task in intelligent transportation systems, which has
a wide range of applications. In this paper, we propose to
integrate vehicle detection and vehicle classification into
one single framework by using deformable part-based mod-
els. First of all, we use annotated vehicle images to train a
deformable part-based model for each class of vehicles to
be classified. Then, given a traffic scene image, we employ
the obtained vehicle models to perform vehicle detection in
it for vehicle extraction. After that, model alignment is per-
formed on the extracted image crop, based on which features
are extracted for creating a representation for the vehicle in
the given image. We train a linear multi-class Support Vec-
tor Machine classifier based on representations of images
in a validation set. Finally, we adopt the SVM classifier for
vehicle classification. The proposed method is evaluated on
the BIT-Vehicle Dataset, and can achieve an accuracy of
91.08%, which is superior to methods used for comparison.
Obtained results demonstrated the effectiveness of the pro-
posed method.
Keywords Vehicle detection · Vehicle classification ·
Deformable part-based model · Support Vector Machine ·
Appearance feature
B
Chang Yao
cyao@m.bjtu.edu.cn
1
School of Electronic and Information Engineering, Beijing
Jiaotong University, No. 3 Shang Yuan Cun, Hai Dian
District, Beijing, China
2
Information Center, National Natural Science Foundation of
China, Beijing, China
1 Introduction
In order to obtain qualified intelligent transportation systems
(ITS), varieties of tasks need to be accomplished, such as
accident detection, congestion control, traffic flow estima-
tion, law enforcement. Because vehicle classification enables
ITS to identify behaviors of different types of vehicles, it
plays an important role and has a wide range of applications
in ITS [3].
Although sensors like radar, or magnetic induction can
be utilized to realize vehicle classification, they are either
intrusive or need high maintenance costs, which limits their
use. Contrarily, as cameras are used extensively for traffic
surveillance, more and more researchers rely on vision-based
systems for solving various kinds of problems, making it one
of the most important research areas in ITS [15]. However,
even though it is relatively easy for a human being to identify
the class of a vehicle in an image, accomplishing this task is
still an open question for a computer [12].
So far, in order to recognize the classes of vehicles in traf-
fic scene images, many approaches have been proposed in
the computer vision community [1], including methods that
explore vehicle’s 3D parameters [10,26] and methods that
utilize vehicle’s appearance features [11,21]. Furthermore,
work on vehicle classification can either focus on vehicle
side view images [16,21] or vehicle frontal view images [3].
Since nowadays traffic surveillance cameras are capturing
increasing numbers of vehicle frontal view images, in this
paper we focus on frontal view vehicle classification. More-
over, because 3D parameters are less reliable for frontal view
vehicle classification, we adopt an approach based on appear-
ances of vehicles.
Specifically, in this paper, to classify frontal view images
of vehicles, we propose an approach which explores
deformable part-based model. The deformable part-based
123